3abiton

42 ms·

1. 871.▲
   by 3abiton 3y ago

   This is an interesting premise that should pursuit further testing on base models rather than instruction-enhanced ones. The assumption that LLMs with all its knowledge is able to ballpark a potentially quantifiable guess is not illogical.
2. 872.▲
   by 3abiton 3y ago

   Has there been any quantified benchmark on this?
3. 873.▲
   by 3abiton 3y ago

   There was also the meme robot some years ago that would reply by gifs. I have no idea what happened to it.
4. 874.▲
   by 3abiton 3y ago

   When did efficiency started to matter with home servers? It feels like chasing the high perfmance/w ratio is a recent thing that is becoming more popular, or maybe I'm just out of the loop. I'm not hating btw, just curious.
5. 875.▲
   by 3abiton 3y ago

   That was an amazing read. Learned quite a lot from your journey! I wish such simple things didn't need much user hacking.
6. 876.▲
   by 3abiton 3y ago

   But this seems to require access to a different oauth token already, ie the victim could have had their tokens from other websites leaked.
7. 877.▲
   by 3abiton 3y ago

   I wonder how the perfomance fair when context size is increased. Intuitively this should be higher, but some quantized models I've tested showed noticeably worst performance.
8. 878.▲
   by 3abiton 3y ago

   Did you stumble upon any resources discussing the history of embeddings and its use in CS and LLMs? It's becoming a cornerstone of ML.
9. 879.▲
   by 3abiton 3y ago

   I'm confuses, reconcialiation is also used in Bayesian statistics. Is it the same context?
10. 880.▲
    by 3abiton 3y ago

    There used to be this video explaining each layer of a neural network visually on youtube back in the days.
11. 881.▲
    by 3abiton 3y ago

    Now with LLM, that might not be far fetched.
12. 882.▲
    by 3abiton 3y ago

    Any possibility that it would hit the mass-market?
13. 883.▲
    by 3abiton 3y ago

    There is also Phind and WizardLM, but I found their performance to be quite poor tbh. It's hard to match gpt-3.5 quality.
14. 884.▲
    by 3abiton 3y ago

    This is the first multimodal model i hear about that is open source. Are there already other alternatives?
15. 885.▲
    by 3abiton 3y ago

    Can't wait till Llemma 4bit
16. 886.▲
    by 3abiton 3y ago

    Isn't ggml supposed to be running on CPU? It seems that this can run over any GPU, not necessarily just NVIDIA
17. 887.▲
    by 3abiton 3y ago

    How does this compare to other stat libraries?
18. 888.▲
    by 3abiton 3y ago

    Finding a good PI is the real advice.
19. 889.▲
    by 3abiton 3y ago

    Let me input that into chatgpt
20. 890.▲
    by 3abiton 3y ago

    Will be used for the next version
21. 891.▲
    by 3abiton 3y ago

    Is this the same as pause tokens?
22. 892.▲
    by 3abiton 3y ago

    I hope one day there will a piratebay for datasets (the pile) and ai models (llama)
23. 893.▲
    by 3abiton 3y ago

    This is awesome, is there something similar tor Android?
24. 894.▲
    by 3abiton 3y ago

    Can't wait for the github repo adaptation of the method!
25. 895.▲
    by 3abiton 3y ago

    Same thing with LoRa. Existing acronyms or not, it's all for grabs.
26. 896.▲
    by 3abiton 3y ago

    Meta research has been really putting great stuff out there. No wonder they are still rocking after all those years, despite the reputation facebook has been racking.
27. 897.▲
    by 3abiton 3y ago

    Is he a moral relativist?
28. 898.▲
    by 3abiton 3y ago

    I think Skiena had a good course on it
29. 899.▲
    by 3abiton 3y ago

    Culprit: non-password protected instances
30. 900.▲
    by 3abiton 3y ago

    At some point Open-source models' performance will follow. Llava will improve.